Is bitstring support still useful?

Is bitstring support still useful?

[Daniel Jacobowitz]

GDB contains both type (TYPE_CODE_BITSTRING) and expression
(OP_BITSTRING) support for bitstrings.  This is in the way for some
bitfield changes I'd like to make, and before I spend time fixing it,
I'd like to know if it's still useful.

OP_BITSTRING is currently dead code.  It's handled in a number of
places, but never generated.

TYPE_CODE_BITSTRING can be generated two ways: from OP_BITSTRING (so,
dead) or from the stabs reader.  It's not generated from any other
symbol reader, including DWARF.  Does this mean it's dead or that
there's something the DWARF reader should do?

For now, I'm going to remove support in my local tree.

Pierre, I've copied you because the most recent non-mechanical patch
related to bitstrings was yours, a long while ago:

2002-05-02  Pierre Muller  <muller@ics.u-strasbg.fr>

        * p-typeprint.c (pascal_type_print_base): Add support
        for TYPE_CODE_STRING and TYPE_CODE_BITSTRING.

(I'm not sure you'll receive my mail directly, though - something
about the IPv6 addresses used by u-strasbg.fr causes my mail server a
lot of grief.)

-- 
Daniel Jacobowitz
CodeSourcery

Re: Is bitstring support still useful?

[Daniel Jacobowitz]

On Fri, Jul 17, 2009 at 10:33:15AM -0400, Daniel Jacobowitz wrote:
> For now, I'm going to remove support in my local tree.

I'd better not... I just realized that I have a potential use for
this, in displaying packed bit arrays found in some device peripheral
registers.

I'd still like to know whether the existing support is used by any
language, though.  If not, we can clean it up aggressively and remove
the stabs reader support for bitstrings.  I suspect that it is mostly
copied and pasted from the removed Chill support.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Is bitstring support still useful?

[Jonas Maebe]

On 17 Jul 2009, at 16:41, Daniel Jacobowitz wrote:

> I'd still like to know whether the existing support is used by any
> language, though.  If not, we can clean it up aggressively and remove
> the stabs reader support for bitstrings.  I suspect that it is mostly
> copied and pasted from the removed Chill support.

The Free Pascal Compiler uses it for bitpacked arrays (arrays whose  
element sizes are not multiples of 8 bits), because there does not  
appear to be any other way to represent them in stabs. I've also tried  
@P, but that seemed to be ignored by gdb (and there is no reason why  
it would mean bitpacked rather than e.g. bytepacked).

Bitpacked arrays are not used very much though, so if you can suggest  
another way to generate Stabs debug information for them, that's fine  
too (i.e., backwards compatibility is not that important in our case).


Jonas

Re: Is bitstring support still useful?

[Daniel Jacobowitz]

On Fri, Jul 17, 2009 at 05:00:54PM +0200, Jonas Maebe wrote:
> The Free Pascal Compiler uses it for bitpacked arrays (arrays whose
> element sizes are not multiples of 8 bits), because there does not
> appear to be any other way to represent them in stabs. I've also
> tried @P, but that seemed to be ignored by gdb (and there is no
> reason why it would mean bitpacked rather than e.g. bytepacked).
> 
> Bitpacked arrays are not used very much though, so if you can suggest
> another way to generate Stabs debug information for them, that's fine
> too (i.e., backwards compatibility is not that important in our
> case).

Does this compiler support DWARF, and if so, how do you represent them
there?

-- 
Daniel Jacobowitz
CodeSourcery

Re: Is bitstring support still useful?

[Jonas Maebe]

On 18 Jul 2009, at 06:09, Daniel Jacobowitz wrote:

> Does this compiler support DWARF,

Yes.

> and if so, how do you represent them
> there?

type
   ta = bitpacked array[1..10] of 0..2;

becomes:

0x0000007d:     TAG_array_type [5] *
                  AT_name( "TA" )
                  AT_byte_size( 0x00000003 )
                  AT_bit_stride( 0x00000002 )
                  AT_type( {0x00000094} ( subrange BYTE ) )

0x00000087:         TAG_subrange_type [6]
                      AT_lower_bound( 0x00000001 )
                      AT_upper_bound( 0x0000000a )
                      AT_type( {0x000000a0} ( SHORTINT ) )

0x0000008e:         NULL


gdb does not handle this yet though (DW_AT_bit_stride is unsupported  
in dwarf2read.c). Printing a variable of this type therefore prints it  
as if it were an array of 10 bytes. But at least the concept is  
representable in DWARF-3, and it just needs fixes in the DWARF reader  
(and possibly in the Pascal printer).


Jonas